defs.rs: async conversion, remove block_in_place

Convert resolve(), resolve_placeholders(), run_agent() to async.
Use memory_render/memory_query directly with .await instead of
block_in_place wrappers.

Propagate async to callers:
- config.rs: resolve(), load_session(), reload_for_model()
- identity.rs: load_memory_files(), assemble_context_message()
- oneshot.rs: run_one_agent()
- prompts.rs: agent_prompt()

Co-Authored-By: Proof of Concept <poc@bcachefs.org>

src/mind/identity.rs

@@ -7,6 +7,7 @@
 use anyhow::Result;
 use std::path::{Path, PathBuf};
 
+use crate::agent::tools::memory::memory_render;
 use crate::config::{ContextGroup, ContextSource};
 
 /// Read a file if it exists and is non-empty.
@@ -71,7 +72,7 @@ fn find_context_files(cwd: &Path, prompt_file: &str) -> Vec<PathBuf> {
 ///   2. Project dir (if set)
 ///   3. Global (~/.consciousness/)
 /// For journal source, loads recent journal entries.
-fn load_memory_files(memory_project: Option<&Path>, context_groups: &[ContextGroup]) -> Vec<(String, String)> {
+async fn load_memory_files(memory_project: Option<&Path>, context_groups: &[ContextGroup]) -> Vec<(String, String)> {
     let home = match dirs::home_dir() {
         Some(h) => h,
         None => return Vec::new(),
@@ -94,12 +95,7 @@ fn load_memory_files(memory_project: Option<&Path>, context_groups: &[ContextGro
             ContextSource::Store => {
                 // Load from the memory graph store via typed API
                 for key in &group.keys {
-                    let content = tokio::task::block_in_place(|| {
-                        tokio::runtime::Handle::current().block_on(
-                            crate::agent::tools::memory::memory_render(None, key, Some(true))
-                        )
-                    });
-                    if let Ok(c) = content {
+                    if let Ok(c) = memory_render(None, key, Some(true)).await {
                         if !c.trim().is_empty() {
                             memories.push((key.clone(), c));
                         }
@@ -141,7 +137,7 @@ fn load_memory_files(memory_project: Option<&Path>, context_groups: &[ContextGro
 }
 
 /// Context message: instruction files + memory files + manifest.
-pub fn assemble_context_message(cwd: &Path, prompt_file: &str, memory_project: Option<&Path>, context_groups: &[ContextGroup]) -> Result<(Vec<(String, String)>, usize, usize)> {
+pub async fn assemble_context_message(cwd: &Path, prompt_file: &str, memory_project: Option<&Path>, context_groups: &[ContextGroup]) -> Result<(Vec<(String, String)>, usize, usize)> {
     let mut parts: Vec<(String, String)> = vec![
         ("Preamble".to_string(),
          "Everything below is already loaded — your identity, instructions, \
@@ -162,7 +158,7 @@ pub fn assemble_context_message(cwd: &Path, prompt_file: &str, memory_project: O
         }
     }
 
-    let memories = load_memory_files(memory_project, context_groups);
+    let memories = load_memory_files(memory_project, context_groups).await;
     let memory_count = memories.len();
     for (name, content) in memories {
         parts.push((name, content));

src/mind/unconscious.rs

@@ -265,7 +265,7 @@ pub async fn prepare_spawn(name: &str, mut auto: AutoAgent) -> Result<SpawnResul
     let exclude: std::collections::HashSet<String> = std::collections::HashSet::new();
     let batch = match defs::run_agent(
         &store, &def, def.count.unwrap_or(5), &exclude,
-    ) {
+    ).await {
         Ok(b) => b,
         Err(e) => {
             dbglog!("[unconscious] {} query failed: {}", name, e);